Improving Soil Properties Mapping and Parameterization for Land Surface Modeling

Abstract

Soil is an important component of the Earth system, influencing the atmosphere, groundwater, and surface processes. Soil properties like texture, structure, and organic matter affect nutrient availability, water movement, and carbon storage, impacting agriculture and hydrology. Accurate representation of soil properties is needed for land surface modeling, guiding agricultural practices, and managing water resources. Earlier modeling approaches relied on soil maps at coarse resolution (> 1-km), often underestimating soil variability. Progress in digital soil mapping (DSM) and initiatives like GlobalSoilMap have improved the availability and quality of georeferenced soil data. Moreover, developments of high-resolution Earth observation data and on-going efforts like the National Cooperative Soil Survey (NCSS) in the U.S. have further contributed to the development of next-generation soil maps.This dissertation focuses on enhancing the spatial representation of soil properties, thereby aiding land surface modeling and agricultural practices. This is achieved through two primary processes: (1) improving soil mapping methodologies and (2) optimizing the utilization of existing soil information. To address data imbalance in soil surveys, we developed a pruned Hierarchical Random Forest (pHRF) method. This framework uses hierarchical soil taxonomy to condition predictions probabilistically and prunes implausible predictions using field-validated samples, reducing uncertainties in estimating soil properties. An iterative bias correction method is further developed to refine prior predictions by adjusting predictive errors. This approach also incorporates additional soil profile data to improve regression trends. These developments establish foundations for creating the Soil Information Refined using Iterative Updated Soil Layers (SIRIUS) dataset, which maps soil properties across six depths (0–2 m) in the Contiguous United States (CONUS). SIRIUS includes soil properties like particle size fractions, pH, bulk density, organic matter, and soil hydraulic properties. SIRIUS outperforms existing datasets, especially in predicting particle size fractions and pH, while maintaining realistic soil property correlations. Although deeper layers have more uncertainty, the framework effectively addresses data imbalances, corrects prediction biases, reduces uncertainties, and improves accuracy in predicting soil properties. Overall, these progresses improve the spatial representation of soil properties, supporting applications in agriculture management and land surface modeling.